Essays on the econometric analysis of structural instabilities and systemic risk

Lee, So Jin

PDF
Thesis_SJLee2025_final.pdf - Veröffentlichte Version
Download (2MB)

URN:	urn:nbn:de:bsz:180-madoc-706240
Dokumenttyp:	Dissertation
Erscheinungsjahr:	2025
Ort der Veröffentlichung:	Mannheim
Hochschule:	University of Mannheim
Gutachter:	Trenkler, Carsten
Datum der mündl. Prüfung:	2025
Sprache der Veröffentlichung:	Englisch
Einrichtung:	Außerfakultäre Einrichtungen > GESS - CDSE (VWL)
Fachgebiet:	330 Wirtschaft
Fachklassifikation:	JEL: C1,C4,C5,
Freie Schlagwörter (Englisch):	factor models , network models , systemic risk , structural breaks
Abstract:	This dissertation consists of three chapters in a study of structural instabilities and systemic risk analysis. A system of interest is assumed to be interconnected, but the underlying structure is unknown or unobservable. Chapter 1 develops a method to detect and localize the points of structural instabilities in the cross-correlation structure of a panel. Cross-correlation structures contain valuable information about the underlying linkages among variables and the channels of spillovers across cross-sectional units. Instabilities in these structures often signal structural changes within the system. We propose a novel method for detecting instabilities in cross-correlation structures using a latent factor model framework. We introduce a suitable object — the column space of the loading matrix (the factor space) — to capture structural correlation changes while being free from the inherent identification issue of the latent model. The resulting detection criterion is based on an intuitive distance measure between two factor spaces, integrating both the detection and localization of breakpoints. In applications, our methods effectively detect instability points consistent with the development of the subprime mortgage crisis, as well as major policy changes such as the repeal of the Glass–Steagall Act and the U.S.–China trade war. Chapter 2 proposes a novel framework to identify the most influential units behind structural breaks. In a system represented by panel data, a break in the cross-correlation structure can empirically indicate volatility propagation from individual (idiosyncratic) dimensions to the entire system. Individual units contributing the most to this break can act as systemic risk components, potentially driving further instability across the system. We propose a novel method to detect these main contributors — referred to as 'granular units'— as an early detection tool for potential systemic risk components. Assuming a standard approximate latent factor structure to model system covariance dynamics agnostically, we introduce a straightforward influence measure to evaluate the contributions of individual (idiosyncratic) second moments to the structural break. Applied to S&P 100 daily return data across major economic crisis periods, the proposed detection scheme effectively identifies likely sources of systemic risk from early crisis stages. Chapter 3 designs a new sequential early warning framework for structural changes that accommodates a broad range of instabilities in the underlying latent network. Network and factor models are two important techniques for analyzing interconnected systems, and we demonstrate that an interconnected system can naturally have a dual representation through our network-factor model. This modeling enables the analysis of instabilities in the latent network using various tools from factor analysis. This online warning framework can be of practical importance for application to network-supported data in which the underlying structure is unknown or unobservable.